The Economic Assimilation of Irish Famine Migrants to the United States
William J. Collins and Ariell Zimran
February 2019
The repeated failure of Ireland's potato crop in the late 1840s led to a major famine and sparked a surge in migration to the US. We build a new dataset of Irish immigrants and their sons by linking males from 1850 to 1880 US census records. For comparison, we also link German and British immigrants, their sons, and males from US native-headed households. We document a decline in the observable human capital of famine-era Irish migrants compared to pre-famine Irish migrants and to other groups in the 1850 census, as well as worse labor market outcomes. The disparity in labor market outcomes persists into the next generation when immigrants’ and natives’ sons are compared in 1880. Nonetheless, we find strong evidence of intergenerational convergence in that famine-era Irish sons experienced a much smaller gap in occupational status in 1880 than their fathers did in 1850. The disparities are even smaller when the Irish children are compared to those from observationally similar native white households. A descriptive analysis of mobility for the children of the famine Irish indicates that having a more Catholic surname and being born in Ireland were associated with less upward mobility. Our results contribute to literatures on immigrant assimilation, refugee migration, and the Age of Mass Migration.
JEL: F22, J61, J62, N31, O15 Keywords: Migration, Refugees, Assimilation, Intergenerational Mobility, Irish Famine
Contact: [email protected]; [email protected]
The authors thank Joseph Ferrie, Tim Hatton, Nick Holtkamp, Joel Mokyr, Cormac Ó Gráda, Marianne Wanamaker; seminar participants at Auburn University, UCLA, the University of Minnesota, Vanderbilt University, and William & Mary; and conference participants at the 2019 ASSA Meeting for helpful suggestions. Claire Whittaker provided excellent research assistance. Collins is the Terence E. Adderley Jr. Professor of Economics at Vanderbilt University and a Research Associate of the NBER. Zimran is an Assistant Professor of Economics at Vanderbilt University and a Faculty Research Fellow of the NBER. 1. Introduction Ireland’s Great Famine in the late 1840s marked a turning point in the country’s demographic and economic history. In 1841, the population numbered just over 8 million; it is estimated that the famine caused the death of about one million Irish and drove another million to emigrate by the early 1850s (Ó Gráda 1999, chapter 3).1 Most of the Irish emigrants settled in the United States, where virtually open borders gave sanctuary from the horrors of starvation and disease (Ó Gráda 2019). The US ultimately absorbed over 500,000 new arrivals from Ireland between the famine’s onset in 1846 and 1850 (Ferenczi and Wilcox 1929; Barde, Carter, and Sutch 2006).2 This paper picks up the emigrants’ story on American shores and builds new datasets to study their and their children’s labor market outcomes in comparison to those of US natives and immigrants from other countries. Irish migration to the US had been growing since the 1830s as part of a general rise in transatlantic migration (Mokyr and Ó Gráda 1982; Ó Gráda 1983; Cohn 2009); but the sharp increase in arrivals during the famine dwarfed previous arrival cohorts. The unprecedentedly large stream of arrivals induced by the famine marked the start of the “Age of Mass Migration,” and likely comprised the largest group of refugees from a single source that the US has ever absorbed relative to the size of its population.3 This change in the volume of Irish immigrants coincided with an apparent change in their characteristics relative to earlier, more prosperous Irish immigrants (Handlin 1991 [1941] p. 51; Miller 1985 p. 295; Anbinder 1992, p. 7), and exacerbated the view of many Americans at the time that the Irish migrants’ relative poverty, tendency to live near one another, and predominantly Catholic religion were barriers to their assimilation and, therefore, justified restrictions on immigration and immigrants’ rights (Anbinder 1992; Hirota 2017). Such concerns have resurfaced throughout US history in response to the arrival of poor or culturally “different” immigrants. In this case, the nativist response was severe, culminating in the political ascent of the “Know Nothings” (Anbinder 1992; Alsan, Eriksson, and Niemesh 2019). This paper studies the economic status and labor market assimilation of the famine-era Irish immigrants, with a particular focus on the adult labor market outcomes of the immigrants’ children. Although the group’s size and historical prominence makes them particularly interesting to study,
1 Ó Gráda (1999, ch. 3) discusses the dating of the famine and the challenges of quantifying deaths and emigration. 2 Cohn (2009, pp. 26-27) describes difficulties in determining the number of Irish immigrants from official US immigration statistics before 1850. 3 For perspective, refugees and asylees granted permanent resident status under the Displaced Persons Act (1948) averaged about 118,000 per year from 1950-52, at the program’s peak (Barde, Carter, and Sutch 2006, p. 1-632); the Mariel Boatlift entailed approximately 125,000 Cuban migrants in 1980 (Card 1990, p. 245).
1 data constraints have made it difficult to do so. To overcome these constraints, we constructed a new micro-level dataset by linking males born in Ireland, Germany, Britain, and the US from the 1850 to the 1880 complete-count US censuses.4 The linked dataset is large, national in scope, and conservatively constructed to reduce false matches. It includes household heads and their sons; therefore, it allows us to compare labor market outcomes for immigrant and native groups over two generations, and to consider the second generation’s labor market outcomes in light of variation in their early life circumstances. Moreover, the new dataset’s panel structure helps us to avoid biases from cohort quality changes and selective return migration that confound inferences from cross- sectional census data (Lubotsky 2007; Abramitzky, Boustan, and Eriksson 2014). That is, because we follow a fixed set of men over time, there are no changes in sample composition. The inclusion of German and British immigrants, who comprised the next largest groups of immigrants in this period, allows for useful comparisons across arrival cohorts and across immigrant groups, which in turn helps illustrate the distinctiveness of the famine-era Irish. The children’s economic outcomes are especially interesting in the context of concerns about the long-run assimilation of new immigrant groups. Immigrants’ children’s outcomes have garnered considerable attention in the literature on the economics of migration (e.g., Borjas 1992, 1993; Card, DiNardo, and Estes 2000; Card 2005; Caponi 2011; Abramitzky, Boustan, and Eriksson 2014; Alexander and Ward 2018). But to our knowledge, this is the first paper to study the children of the large cohorts of immigrants who arrived in the mid-nineteenth century US, at the start of the Age of Mass Migration. This is also one of a small number of papers that address long-run patterns of economic assimilation by refugee immigrants or their children, whose experiences and outcomes may differ from those of other migrant groups (Edin, Fredriksson, and Åslund 2003; Cortes 2004; Beaman 2012; Evans and Fitzgerald 2017).5 Our focus on the immigrants’ children is also a practical consequence of the historical census data’s limitations. The censuses of this era did not inquire directly about each immigrant’s year of arrival. This poses a major challenge to discerning between those who arrived before or after the Irish famine’s onset and, in general, to any study of immigrant assimilation in this early period.6
4 Linking females from childhood to adulthood is difficult due to name changes at marriage. 5 We use the term “refugee” here because a large number of the Irish migrants were permanently driven from home by a severe ecological disaster and ensuing economic and social dislocation in the late 1840s. The modern literature focuses on refugees that meet a particular legal definition established in the postwar period; see Hatton (2012) for a brief review. 6 One reason much of the recent literature on the Age of Mass Migration focuses on the early twentieth century is that the census microdata report year of immigration starting in 1900. Before 1850, the census did not
2 Nonetheless, in 1850, we can determine the arrival cohort for many household heads by examining their children’s birth year and birth place information.7 This approach definitively categorizes the household head’s arrival cohort (“pre-famine” or “famine-era”) for about two-thirds of the sons of Irish immigrants.8 The relatively high classification rate makes studying the children’s outcomes attractive—we know whether their household head arrived before or during the famine, observe their childhood household characteristics in 1850 in detail, and then see their labor market outcomes as prime-aged adult workers in 1880. In addition, the “classified” set of children is fairly representative of all immigrants’ children circa 1850 in terms of observable characteristics. We use the new dataset to address three main sets of questions. First, in 1850, how different were the households headed by famine-era Irish migrants from those headed by earlier Irish migrants, by concurrent migrants from Germany and Britain, and by US natives? In particular, is there evidence of a differential change in Irish household heads’ human capital (reflecting changing selection) and labor market outcomes (reflecting both selection and labor market conditions for newly arrived migrants) between the pre-famine and famine-era migrants? Given this paper’s motivation, we care about changing migrant selection primarily because it directly influences the average household characteristics and resources of immigrants’ children, though documenting the patterns of migrant selection at the time of the famine is also of independent interest (e.g., Mokyr and Ó Gráda 1982; Cohn 1995). Consistent with the predictions of a simple Roy (1951)-Borjas (1987) model, we find clear evidence of deterioration in the human capital of Irish immigrant household heads with the onset of the famine, as measured by literacy and age heaping in 1850. This is the clearest evidence to date on human capital differences between pre-famine and famine-era arrivals.9 We also document a decline in occupational status between the pre-famine and famine-era Irish, both absolutely and relative to differences over arrival cohorts for other immigrant groups. Thus, by changing both the composition and volume of Irish migration, the Great Famine resulted in a large inquire about place of birth; so, there is limited scope for backward linkage. Extending links to the passenger lists in this early period is difficult due to the sparseness of records and the feasibility of making multiple accurate links across various datasets (see footnote 30). 7 For example, a family with an Irish-born child aged 2 in 1850 immigrated to the US some time in or after 1848, which would make them famine-era migrants. A family with an American-born child aged 7 in 1850 immigrated to the US some time before or during 1843, making them pre-famine migrants. We make the same distinctions for the British and German migrants to form comparison groups of arrival cohorts. 8 In contrast, this approach enables us to classify the arrival cohort of fewer than half of adults in the 1850 census. The method proves to be very accurate in cases where the classification can be verified with outside information (i.e., searching passenger lists). 9 This contribution to the literature on the Irish famine is limited by the fact that our method to classify individuals’ year of arrival requires that they have children; this group may have differed meaningfully from childless immigrants.
3 cohort of relatively poor immigrant children in the US circa 1850. Second, in 1880, how did the adult labor market outcomes of Irish famine migrants’ sons compare to those of other immigrants’ and natives’ sons? Studying these outcomes provides perspective on the potential for long-term assimilation in a setting where a variety of forces— including inauspicious early life conditions, an environment rife with anti-Irish sentiment, and a continuing influx of new immigrants—may have hindered the Irish children’s advancement. We find that, on average, the sons of the famine-era Irish immigrants fared poorly in the labor market as adults in comparison with other groups in 1880. Nonetheless, in comparison to their fathers’ starting point, they greatly narrowed the gap in occupational status relative to natives. The gap is even smaller when considering their adverse childhood environment (i.e., when controlling for 1850 household characteristics); that is, the intergenerational occupational upgrading of the famine Irish was almost the same as that of observationally similar natives. In this sense, fairly strong economic assimilation occurred over generations. The labor market outcomes of Irish immigrants circa 1850 were thus a poor guide to inferring the group’s ability to assimilate over a longer period. Finally, we examine heterogeneity in the upward mobility of the children of famine-era Irish immigrants according to observable characteristics, including measures of their father’s human capital, residence in an Irish enclave, geographic mobility, and social distance (specifically, having a surname that we determine to be predominantly Catholic). The data and setting do not allow clear identification of causal relationships; they do, however, provide novel evidence on factors that may have facilitated or impeded immigrants’ intergenerational gains. In particular, we find that having a more Catholic surname and being born in Ireland were associated with less upward mobility, conditional on other observables. This pattern is consistent with the presence of discrimination against Irish Catholics, lower levels of human capital for Irish Catholics, and long-term negative consequences from exposure to the famine, all of which merit closer examination in future research. This paper contributes to the literature on the economics of immigration in several dimensions. Most directly, the paper advances our knowledge of the Age of Mass Migration generally and of Irish migration during the famine specifically. It is thus complementary to research by Hatton and Williamson (1998), Ferrie (1999), Cohn (2009), and Abramitzky, Boustan, and Eriksson (2012, 2014), among others. By studying immigrants’ experiences in the early portion of this period, in contrast to a recent scholarship that has focused on the post-1900 period, the paper brings the first cohorts of the Age of Mass Migration into sharper focus. More broadly, studying the experience of Irish famine-era migrants may yield broader insights into the economics of large-scale
4 migration due to natural or man-made disasters.10 By studying a large wave of migrants and their offspring long before the implementation of restrictive immigration policies, the paper yields insight into the migration process and potential for assimilation in a setting where virtually open borders and lack of labor, housing, or other regulation allowed economic forces to predominate. Finally, our emphasis on economic assimilation in the long run speaks to a core theme of the international migration literature, both historical and contemporary.11 Issues of migrant assimilation have become particularly salient in recent years, as large numbers of refugees have sought residence in the United States and Europe (United Nations High Commissioner for Refugees 2017). This has fueled a debate in which concerns about immigrants’ assimilation have been cited as justification for more restrictive policies (e.g., Kelly 2018), such as proposals that potential entrants be screened to favor those with a greater “likelihood of successful assimilation and contribution to the United States” (US Department of State et al. 2017, p. 8). In the Irish case, we find that despite a notable deterioration in migrant selection and a significant political backlash, the migrants’ children converged strongly, albeit incompletely, on natives’ outcomes by 1880. In this sense, they showed clear evidence of “assimilation and contribution” to the American economy.
2. Background on Ireland’s Great Famine and Migration to the US On the eve of the Great Famine, two-thirds of Irish families were employed primarily in agriculture (Commissioners 1843, p. xviii). Most owned little or no land (Ó Gráda 1999, p. 25) and had few financial resources. Widespread poverty and heavy reliance on the potato left the Irish vulnerable to large and repeated failures of the potato crop, as occurred in 1845 and 1846 due to the spread of a microorganism that causes blight (Phytophthora infestans). By late 1846, Ireland was in the grips of a historic famine. The relatively good yields of 1847 were of little avail as farmers had shifted away from planting potatoes. The famine was then exacerbated by crop failures in 1848 and, less severely, in 1849 and 1850.12 Even before the famine, the migration flow from Ireland to the US was substantial (Mokyr
10 On migration due to environmental disasters or war see recent work by Boustan, Kahn, and Rhode (2012), Hornbeck (2012), Moser, Voena, and Waldinger (2014), Mahajan and Yang (2017), Nakamura, Sigurdsson, and Steinsson (2017), Blum and Rei (2018), and Long and Siu (2018). 11 On immigrant assimilation, both historical and contemporary, see inter alia Ferenczi and Wilcox (1929), Chiswick (1978), Borjas (1992, 1993), Hatton (1997), Hatton and Williamson (1998), Card, DiNardo, and Estes (2000), Minns (2000), Cortes (2004), Card (2005), Abramitzky, Boustan, and Eriksson (2014, 2016), and Alexander and Ward (2018). 12 Mokyr (1983) and Ó Gráda (1999) discuss the chronology of agricultural failure and famine in detail. See Solar (1989) on Irish crop failures in historical and international perspective.
5 and Ó Gráda 1982). Historians suggest that pre-famine migrants were likely drawn from above the Irish median in terms of skill and socioeconomic status, but perhaps less so over time as costs of migration declined (Miller 1985, pp. 193-201). Consistent with the impression of positive selection, whereas approximately 40 percent of men (ages 16-45) in Ireland could neither read nor write in 1841 (Commissioners 1843, p. xxxvi), only 15 percent of Irish-born men (ages 25-54) were classified as illiterate in the 1850 US census.13 The bulk of the Irish famine emigrants, including many of those who originally landed in Canada, settled in the US (McInnis 2000; Cohn 2009). Passenger records from US ports of entry suggest that the annual rate of Irish immigration nearly doubled with the famine’s onset.14 By June 1850, when the US population census was taken, there were nearly one million Irish-born residents. To be sure, some of those who migrated during the famine would have left Ireland in any case. But Ó Gráda points out that “… most of the mass emigration of the late 1840s was part of the famine tragedy. It was push migration with a vengeance, and its tragic character has been rightly at the center of historical writings about it” (1999, p. 105). Miller writes of a “mass, indiscriminate rush to leave” Ireland in the wake of the potato blight (1985, p. 292). Many of the migrants were making a leap of faith in a desperate context characterized by widespread death and social collapse. Like modern day refugees, their departure was precipitated by an imminent threat to survival, often undertaken hastily, and generally irreversible.15 A major difference, of course, is that migrants fleeing to the US during the famine did not face the legal restrictions on entry that are prevalent today. In addition to generating a surge in the volume of migration, there are indications that the famine may have led to a change in the composition of Irish migration to the US. Some clues come from fragmentary information on the migrants’ places of origin. It appears that the main sources shifted somewhat from the North and East (Ulster and Leinster) toward the South and the West (Munster and Connacht) (Ó Gráda 1999, p. 113; Anbinder and McCaffrey 2015; Anbinder, Ó Gráda,
13 These are our calculations from 1850 complete count census. We caution that comparing across the censuses this way is somewhat tenuous; it is unclear how comparable literacy measures are. The differences are sufficiently large in this case that we think the information is useful. See Connor (2019) for evidence of negative selection in the early twentieth century. 14 Cohn (2009, pp. 26-27) warns that the time-series data on Irish immigration in the early to mid-nineteenth century allocates a fixed share of UK immigrants to Ireland (0.70) rather than providing a true count. 15 The post-famine transformation of the Irish economy, which included consolidation of small landholdings, ensured that emigration continued once famine conditions passed (Guinnane 1997; Ó Gráda 1999). The parallel between the famine migrants and modern refugees fleeing war and instability has been noted in the Irish press (Delaney 2015).
6 and Wegge 2017). This suggests that the characteristics of migrants may have changed because residents of the South and West were more agricultural and less literate than elsewhere (Commissioners 1843, Geary 1996); they were also more likely to be Catholic and to speak Irish. Passenger lists of ships entering US ports are also potentially informative about changing selection. The lists include self-reported occupation and age, typically recorded at the European port of embarkation. They provide a window on the pre-departure activity of migrants, but the evidence is ambiguous and much caution is necessary (Cohn 2009, ch. 5), which is one factor that motivates our census-based approach.16 Cohn (1995) finds a slight decrease in the fraction of Irish immigrants who were reported as laborers after 1845, whereas Mokyr and Ó Gráda (1982) find an increase in innumeracy as implied by age heaping in passenger lists after 1845. In the US, Irish immigrants were over-represented in urban areas of the Northeast. It is worth emphasizing, however, that most Irish immigrants, including those we identify as famine-era arrivals, did not reside in New York City (13.8 percent of all Irish-born) or Boston (3.5 percent) in 1850.17 Landmarks in the historical literature tend to focus on these cities (Handlin 1991 [1941], Ernst 1949), but a much broader geographic scope is required to describe the experience of the Irish immigrants. Scholars have long understood that Irish immigrants were disproportionately employed in relatively low-skill, low-paying lines of work (Handlin 1991 [1941]; Ernst 1949; Miller 1985). Ferrie (1994, 1995, 1997a, 1999) provides a more dynamic view by drawing a sample of immigrant men from New York City’s passenger lists in the 1840s and linking them to census records in 1850 or 1860. He finds a great deal of mobility between pre-migration occupational categories, coded from passenger lists, and those reported in the 1850 or 1860 US census (1999, Table 5-2). Irish men had lower pre-migration occupational status than the British and Germans based on the passenger lists (i.e., more unskilled laborers), had lower rates of upward mobility from the unskilled category, and accumulated less real estate wealth (1999, Table 6-4). Even so, Anbinder, Ó Gráda, and Wegge (2017) find evidence of upward mobility in the account records of Irish immigrants at the Emigrant Industrial Savings Bank in New York. In addition to their economic struggles, mid-century Irish immigrants were subject to a sharp nativist backlash that was rooted in long-standing anti-Catholic sentiment in the US and heightened
16 The lists are potentially incomplete in coverage, may have deteriorated in quality during the famine, and may have varied in enumeration practices across ports of embarkation. Even within a given list, clerks may or may not have accurately recorded each passenger’s primary past occupation. 17 These are our calculations from the complete count census data (Minnesota Population Center 2017; Ruggles et al. 2015).
7 by the sheer volume of mass migration.18 Hirota (2017, p. 2) writes, “The newcomers’ religion triggered an outburst of anti-Irish nativism in these states [New York and Massachusetts], but so too did the immigrants’ poverty.” Hirota (2017, ch. 2) documents cases of deportation and efforts to prevent destitute Irish passengers from disembarking in Massachusetts. A salient manifestation of this backlash was the success of the “Know Nothing” party in the early 1850s, which advanced the ideas that “Protestantism defined American society” and that “Catholicism was not compatible with the basic values Americans cherished most” (Anbinder 1992, p. 104). Despite much valuable scholarship on Irish immigrants in the US, a number of important gaps remain in our understanding of the famine migrants’ experience. First, there is no clear picture of differences between the Irish who arrived before the famine and those who arrived during the famine. To date, the evidence is somewhat fragmentary and inconclusive. Moreover, although scholars have explored the economic assimilation of first-generation Irish immigrants (e.g., Ferrie 1999), there is no information available on the assimilation experience of the immigrants’ children. This absence stands out relative to studies of modern migrant assimilation (e.g., Borjas 1993; Card 2005), in which children’s outcomes are of fundamental interest. Finally, the economic mobility of the famine Irish has not been closely compared to that of natives, let alone in an intergenerational framework that allows us to tie children’s outcomes to their early life circumstances at an individual level. The mid-nineteenth century was a period of high intergenerational mobility among US natives (Long and Ferrie 2013). It is thus essential to compare immigrants and natives to determine whether immigrants closed the gap in economic status and, in that sense, assimilated over generations. This paper develops new evidence on all of these fronts.
3. Data sources and construction To construct the dataset, we first obtained the complete count US census data for 1850 (Ruggles et al. 2015; Minnesota Population Center 2017). Because this census (and all those prior to 1890) did not inquire about immigrants’ year of arrival in the US, our first task was to develop an alternative approach to distinguish the Irish who arrived during the famine from those who arrived prior to it. We determined the arrival cohort for each Irish-born head of household based on the ages and places of birth of children in the household in the 1850 census.19 Household heads with a child
18 By this time, Irish immigrants were overwhelmingly Catholic. Many German immigrants were also Catholic, but they were better off and less concentrated in cities of the Northeast than the Irish. 19 We use an indicator for relationship to the head of household provided in the original data by Ruggles et al. (2015). Our understanding is that individuals are assumed to be children of the head of household if they are of
8 (son or daughter) born in the US in 1845 or earlier are categorized in the pre-famine group, while those with a child born in Ireland in 1846 or later are in the famine-era group.20 We performed a similar classification for British- and German-born heads of household in 1850.21 We include the Germans and British in this study because, after the Irish, they were the largest groups of immigrants before 1850 (Barde, Carter, and Sutch 2006, p. 1-560), and they provide a useful point of comparison. This classification scheme leads to six categories of immigrant household heads: pre- famine Irish, famine-era Irish, and similar cohorts for the British and Germans. We also include households headed by native-born white men for comparison. We assigned the household heads’ arrival cohort classification and ethnicity to their sons aged 18 or younger. For example, sons of an Irish father who arrived in the US between 1846 and 1850 are referred to as members of the “famine-era Irish.”22 Importantly, not all children in Irish, British, or German-headed households were foreign-born. It follows that not all children classified as being famine-era Irish would have experienced the famine in Ireland (though 40 percent of our benchmark sample of famine Irish sons were born in Ireland during the famine). Given that the paper’s main goal is to characterize long-term assimilation as revealed in the outcomes of immigrants’ children, partitioning the data this way provides simple but new insights. Table 1 divides each immigrant group using the categorization described above, reporting for heads of household in columns (1) to (3) and for sons in columns (4) to (6). The fraction of individuals categorized definitely into the pre-famine or famine-era cohorts was approximately 65 percent for the Irish sons, but only about 44 percent for all Irish heads of household (including those without children).23 The Germans and British exhibit similar patterns. As mentioned earlier, the greater ability to categorize children than heads of households or other adults is part of our motivation for focusing on the children’s long-term outcomes: the children who we can categorize
appropriate age and if they share a surname with the head of household. We must rely on this imputed indicator because the census did not inquire regarding relationship to the head of household until 1880. 20 It is not possible to place all household heads into one of these two categories, and therefore we create two more categories: those with a child born in the United States in 1846 or later and a child born in Ireland between 1841 and 1845 (“probable famine-era”), and all others (“uncategorized”). We do not use these latter two groups in our main analysis. The outcomes of the “probable” group resemble those of the “famine” group, while those of the “uncategorized” group fall between those of the “famine” and “pre-famine” groups. 21 The data obtained from Ruggles et al. (2015) combine England, Scotland, and Wales into a single birthplace of Great Britain. We retain this combination. 22 In principle it is possible to determine for a subset of the sample whether individuals arrived in the early period of the famine (when it was somewhat less severe) or later. But pinning down arrival dates this finely would limit the sample to those with children born in quick succession, severely limiting the sample size. 23 The main group that we cannot classify by our approach (and who are not included in Table 1) is boarders, who were 27 percent of all Irish-born adult males in the 1850 US census.
9 are fairly representative of the population of immigrants’ children (discussed in more detail below), whereas immigrants with children are not likely to be representative of all adult immigrants. In the next step, we linked all the males who were born in Ireland, Britain, Germany, and the US and observed in the 1850 US census to the full count 1880 US census (Ruggles et al. 2015; Minnesota Population Center 2017). We briefly describe the linking procedure here and provide additional details in Appendix A. We do not constrain the sample for linking to those whose arrival cohort could be determined (described above), as this might result in more false matches. Instead, we attempt to link the 1850 census records to 1880 without regard for whether arrival cohort could be determined; we then use only successful links (described below) whose arrival cohort could be clearly classified as pre-famine or famine-era for the assimilation analyses. At the start of the linkage process, we retained only individuals who were unique in terms of place of birth, age-implied year of birth (plus or minus four years), and first and last name (with allowances for orthographic differences) in 1850. We linked those men to the 1880 census, according to the same identifying characteristics and retained only matches in which both the 1850 and the 1880 record had a unique match in the opposite census. This linkage method is essentially similar to that introduced by Ferrie’s (1996) pioneering effort of mechanical record linkage. Relative to Ferrie’s (1996) approach, our method incorporates recent innovations, such as the replacement of the NYSIIS name standardization algorithm with the use of orthographic distance measures, as in Beach et al. (2016), which reduces the danger of false links and has become common in the literature (Bailey et al. 2017; Abramitzky, Mill, and Pérez 2018). We chose this approach because of its long- standing use and its relatively conservative nature in selecting matches. Table 2 presents rates of successful linkage. Column (1) shows the number of observations with which the linkage process began for each 1850 group. Column (2) shows the fraction of individuals who remained after removing those who were not unique on name, birth place, and age. The fraction of these individuals who were successfully located in 1880 is shown in column (3). Among the Irish children, we linked about 15 percent of the unique famine-era Irish and about 20 percent of the unique pre-famine Irish. Among Germans, the rates of linkage are comparable to the Irish, whereas they are between 23 and 26 percent for the British and Americans. These match rates for the British and Americans are comparable to those of Ferrie (1996, p. 145 and Table 4). The match rates for the Irish and Germans are somewhat lower, and there are reasons to expect this. Besides the lower levels of literacy and education of the Irish, which may tend to reduce match rates, the arrivals of large cohorts of Irish and German immigrants after 1850 would have potentially
10 confounded links even of individuals who were unique in 1850.24 In sum, our benchmark dataset of children consists of males who were aged 0 to 18 in 1850, sons of the household head in 1850 (according to the Ruggles et al. 2015 classification), successfully linked from the 1850 to the 1880 US census, born in the US, Ireland, Britain, or Germany, and with a father born in the US, Ireland, Britain, or Germany. We also require that, if the child’s head of household in 1850 was an immigrant, it was possible to determine whether the head of household was in the famine-era or the pre-famine arrival cohort.
Potential pitfalls in data construction Our data construction process is subject to four main potential pitfalls. First, only a fraction of 1850 individuals are linked to the 1880 census, and selection into linkage is likely to be non- random. In Appendix A, we present linear probability models that relate the probability of being linked to a variety of characteristics in 1850. There are several statistically significant predictors of selection into linkage, such as higher property ownership and literacy; this is typical in studies that link census data (e.g., Abramitzky, Boustan, and Eriksson 2014; Beach et al. 2016). However, most differences are small, as indicated in Figure 1, which compares means of variables for linked children and all children of Irish migrants in 1850. To address selection into linkage, we weight all of the analyses by the inverse of the estimated conditional probability of successful linkage.25 The second issue is that some automated linkage methods may yield a high rate of false positives in matching (Bailey et al. 2017). Our linkage approach mitigates this concern, though it cannot be entirely avoided. We eliminate individuals who are not unique in terms of the linkage characteristics from the sample that we attempt to link; this reduces the probability of a making false match. Moreover, we do not make links on the basis of a name standardization algorithm, a practice that Bailey et al. (2017) highlight as a source of false positives.26 To verify that false positives in linkage are not responsible for our results, we have also examined a sample of exact unique matches
24 For instance, a Patrick Kelly born in 1845 in Ireland may have been unique in the United States in 1850, but other Patrick Kellys born in 1845 in Ireland may have arrived after 1850, preventing the identification of the correct Patrick Kelly in 1880. 25 This probability is computed from separate probit regressions of a linkage indicator on a variety of covariates for each of the seven ethnicity-cohort groups. 26 We use the soundex algorithm to reduce the set of candidate matches, but all of the linking is ultimately based on orthographic distance, which is a score derived from the number and type of spelling changes that would be required to move from one name spelling to another.
11 only.27 This further reduces the probability of false matches and does not qualitatively affect our main results. A third concern is that the roughly two-thirds of children whose father’s arrival cohort is definitively classified might not be representative of the population of children in 1850. To determine whether selection into successful classification is likely to bias our results, we compare the means of 1850 observables of the classified Irish children to those of the whole sample of Irish children in Figure 2.28 Though the classification does appear to favor individuals with better socioeconomic status, such as those whose household heads owned more property in 1850, the differences between the classified group and all children are small.29 Finally, there is concern that the categorization of arrival cohort is confounded by incorrect reporting of children’s ages. We view the danger of age heaping as less severe for children than it would be for adults due to the greater developmental differences by age among children. To provide direct evidence, we randomly selected 50 classified households from our linked data, stratifying to include 25 of each categorization group (famine and pre-famine) and searched carefully for their passenger list records using Ancestry.com. Although we were able to locate only 10 of these households definitively,30 all 10 were correctly categorized by the age- and birthplace-based algorithm.
Additional variables of interest The US census did not collect information on income before 1940; therefore, we rely heavily on occupation to indicate men’s relative economic status. The simplest approach is to define and
27 We define an exact match as one in which there is no more than a one-year difference in age-implied birth year and the first and last names of both records are identical after removing double letters and common first name abbreviations (e.g., changing “Wm” to “William”). These results are available on request. 28 Figure 2 covers the linked sample and is weighted to correct for selection into linkage. Similar results are obtained for the children of British and German immigrants. 29 Results are qualitatively unchanged when we re-weight our sample to correct for selection into both linkage and classification on the basis of observables. However, we prefer not to re-weight to correct for selection into classification in our benchmark results because it is not clear whether differences in observables between the classified and unclassified groups reflect selection into classification (which one would like to reverse) or actual differences in the probability that classified and unclassified individuals arrived during the famine. 30 This also illustrates the difficulties of linkage from the passenger lists to the census. Combined with the number of observations lost in the link from 1850 to 1880, requiring an additional link to the passenger lists would result in a very small sample. Even achieving this low rate of linkage to the passenger lists required using information on individuals’ household composition; in some cases, individuals could be located only when searching for their household members. The use of household information in linkage is generally avoided due to the bias it introduces. The census-based approach to determining the arrival year enables us to avoid these problems of linkage from the passenger lists and to maintain a larger sample.
12 study a categorical variable for “unskilled labor,” which combines “laborers” and other relatively low-skilled occupations, such as “porter” and “miner” (following Ferrie 1995, 1999). In addition, we create a variable that exploits the 1870 census’s information on real and personal wealth. For each occupation-by-region (North/South) cell, we calculate a “wealth score” that equals the average total wealth of men (age 30-65) in the one-percent sample of the 1870 census (Ruggles et al. 2015; Minnesota Population Center 2017).31 This is akin to the occupation score variable based on 1950 income data that is often used in studies of twentieth-century US labor markets.32 We believe the 1870-based occupation score is better suited for studying the mid-nineteenth century because it is, of course, closer in time to our period of study, and it is designed to allow for differences between the North and South. We also use this wealth score to define an occupational rank variable, which represents the percentile of the wealth score in the sample.33 For analysis of heterogeneity in outcomes among the Irish (Section 5), we constructed three additional variables. The Irish immigrants’ predominantly Catholic religion was salient and controversial at the time, but of course there was variation in religious affiliation among the Irish. The US population census has never inquired about individuals’ religious affiliation, and no such information is available from passenger lists. To gain some insight on the association between religion and economic outcomes for the Irish, we construct a measure of “Catholicity” based on surnames.34 In particular, we use full-count micro-level data from Ireland’s 1901 census to calculate the fraction of individuals that were Catholic by surname; then, we merge that surname-based information with our linked dataset from the US census.35 The details of the procedure are provided in Appendix B. This adds a useful dimension to the description of Irish immigrants and within-Irish heterogeneity in labor market outcomes. We also create a variable that measures the fraction of the 100 closest households that were
31 We have also experimented with wealth scores at the source country-occupation-region level. Unfortunately, this cuts the data thinly in many instances, and it is not very well suited to our study of immigrants’ children (many of whom were US-born). The 1850 census did not ascertain personal wealth, and the 1880 census did not inquire about wealth at all. 32 The correlation between our measure in 1880 and the 1950-based occupational income score is only 0.28 in our linked data. However, this is largely due to the changing occupational status of farmers in the intervening period. When farmers are excluded, the correlation is 0.75. 33 In the interest of brevity, we do not report the results for occupational rank in all specifications below. Where it is not shown (and where it is), it is qualitatively similar to the results with the wealth score. 34 The use of information contained in surnames to glean additional information on immigrant origins is increasingly common (e.g., Spitzer 2018; Spitzer and Zimran 2018; Pérez 2019b), including in the study of the origins of the Irish in Ireland and abroad (Ferrie 1997c; Ó Gráda 2016; Connor 2019). 35 Ideally, we would have used data from an earlier Irish census. However, the 1901 census is the earliest with micro data available; the 1841 records were destroyed by fire in the 1920s (National Archives of Ireland n.d.).
13 headed by Irish men in 1850, as a way to gauge residence in Irish immigrant enclaves. In this, we follow Logan and Parman (2017) and rely on the idea that individuals close to one another on census manuscript pages were likely geographically proximate to one another. Finally, we created an indicator for having moved states between 1850 and 1880, which may shed light on whether geographic mobility was associated with upward occupational mobility for the Irish. Geographic mobility is a major theme in American economic development, especially in the nineteenth century, but whether it was a direct conduit for Irish immigrants’ economic gains is an open question.
4. Results describing the selection and economic assimilation of Irish immigrants We first describe the 1850 household characteristics of the children of famine-era immigrants. The 1850 data are informative regarding three aspects of the famine-era migration. First, differences in the human capital characteristics of fathers over arrival cohorts (i.e., in their literacy and numeracy) are informative about changes in migrant selection during the famine.36 Second, fathers’ labor market outcomes are informative about the combination of changing migrant selection and difficult labor market conditions experienced by immigrants arriving during the Great Famine. Finally, and most importantly given our focus on intergenerational assimilation, all of the 1850 variables are informative about the early-life conditions experienced by the children of immigrants. Then, we examine the children’s outcomes in the labor market in 1880. By this time, the children are between 30 and 48 years old, a useful range for observing adult labor market outcomes in an intergenerational framework (Haider and Solon 2006, Long and Ferrie 2013, Feigenbaum 2018). Individual-level information on income or wealth is not available in 1880, and so we study occupational status as defined above. We are particularly interested in whether the famine-era Irish children narrowed gaps in status relative to the children of US natives and other immigrant groups. We are also interested in how they fared in a conditional sense—whether they did as well as observationally similar children from other groups. The ability to connect the children’s adult outcomes to their childhood households is a useful feature of the dataset, and it is rare in studies of US immigrant assimilation.
36 There is a concern that the census data do not reflect migrants’ characteristics at time of arrival. Our focus on human capital mitigates this concern because it is less likely that immigrants’ literacy and numeracy changed after arrival than did other measures, such as occupation.
14 Human capital and occupational status of household heads in 1850 Table 3 summarizes the 1850 variables for our main sample of linked and classified children, reported separately by ethnicity and arrival cohort. The first set of variables describes the individual’s father in 1850.37 We collect literacy directly from the census, and we construct a measure of numeracy, which is an indicator for reporting an age that is not divisible by five.38 We characterize occupational status with the wealth score, as described above, and with broad occupational groups. We also observe directly the household head’s real property ownership in 1850. For the children, we observe whether they attended school in the year prior to the census (ages 5 to 15). We also create indicator variables for whether the child was US-born, and in the case of Irish children, whether the child was in utero during the famine (i.e., Irish-born 1846-1850). Differences between Irish and non-Irish households are immediately apparent from these summary statistics, as are differences between pre-famine and famine-era arrival cohorts of Irish immigrants. Relative to natives and to other immigrant groups, both arrival cohorts of the Irish were more likely to be illiterate, innumerate, and to hold unskilled occupations; as a result of the latter, they also had lower occupational wealth scores and ranks. The Irish also owned less property than other similarly tenured immigrants (or natives). Within the Irish, there was a considerable disadvantage for the famine-era arrivals compared to the pre-famine arrivals along these same dimensions. The famine-era Irish household heads were less literate and numerate, were more likely to hold an unskilled occupation, and had more Catholic surnames than earlier Irish immigrants. They were also more likely to reside in an urban area and in proximity to other Irish than were earlier Irish immigrants. To characterize these differences more precisely, we estimate a regression specification of the form